Median Dice Similarity Coefficient Research Articles

Semi-supervised COVID-19 volumetric pulmonary lesion estimation on CT images using probabilistic active contour and CNN segmentation

PurposeA semi-supervised two-step methodology is proposed to obtain a volumetric estimation of COVID-19-related lesions on Computed Tomography (CT) images. MethodsFirst, damaged tissue was segmented from CT images using a probabilistic active contours approach. Second, lung parenchyma was extracted using a previously trained U-Net. Finally, volumetric estimation of COVID-19 lesions was calculated considering the lung parenchyma masks.Our approach was validated using a publicly available dataset containing 20 CT COVID-19 images previously labeled and manually segmented. Then, it was applied to 295 COVID-19 patients CT scans admitted to an intensive care unit. We compared the lesion estimation between deceased and survived patients for high and low-resolution images. ResultsA comparable median Dice similarity coefficient of 0.66 for the 20 validation images was achieved. For the 295 images dataset, results show a significant difference in lesion percentages between deceased and survived patients, with a p-value of 9.1 × 10−4 in low-resolution and 5.1 × 10−5 in high-resolution images. Furthermore, the difference in lesion percentages between high and low-resolution images was 10 % on average. ConclusionThe proposed approach could help estimate the lesion size caused by COVID-19 in CT images and may be considered an alternative to getting a volumetric segmentation for this novel disease without the requirement of large amounts of COVID-19 labeled data to train an artificial intelligence algorithm. The low variation between the estimated percentage of lesions in high and low-resolution CT images suggests that the proposed approach is robust, and it may provide valuable information to differentiate between survived and deceased patients.

Validation of a Fully Automated Hybrid Deep Learning Cardiac Substructure Segmentation Tool for Contouring and Dose Evaluation in Lung Cancer Radiotherapy

Background and purposeAccurate and consistent delineation of cardiac substructures is challenging. The aim of this work was to validate a novel segmentation tool for automatic delineation of cardiac structures and subsequent dose evaluation, with potential application in clinical settings and large-scale radiation-related cardiotoxicity studies. Materials and methodsA recently developed hybrid method for automatic segmentation of 18 cardiac structures, combining deep learning, multi-atlas mapping and geometric segmentation of small challenging substructures, was independently validated on 30 lung cancer cases. These included anatomical and imaging variations, such as tumour abutting heart, lung collapse and metal artefacts. Automatic segmentations were compared with manual contours of the 18 structures using quantitative metrics, including Dice similarity coefficient (DSC), mean distance to agreement (MDA) and dose comparisons. ResultsA comparison of manual and automatic contours across all cases showed a median DSC of 0.75–0.93 and a median MDA of 2.09–3.34 mm for whole heart and chambers. The median MDA for great vessels, coronary arteries, cardiac valves, sinoatrial and atrioventricular conduction nodes was 3.01–8.54 mm. For the 27 cases treated with curative intent (planned target volume dose ≥50 Gy), the median dose difference was –1.12 to 0.57 Gy (absolute difference of 1.13–3.25%) for the mean dose to heart and chambers; and –2.25 to 4.45 Gy (absolute difference of 0.94–6.79%) for the mean dose to substructures. ConclusionThe novel hybrid automatic segmentation tool reported high accuracy and consistency over a validation set with challenging anatomical and imaging variations. This has promising applications in substructure dose calculations of large-scale datasets and for future studies on long-term cardiac toxicity.

Independent Validation of a Deep Learning nnU-Net Tool for Neuroblastoma Detection and Segmentation in MR Images.

To externally validate and assess the accuracy of a previously trained fully automatic nnU-Net CNN algorithm to identify and segment primary neuroblastoma tumors in MR images in a large children cohort. An international multicenter, multivendor imaging repository of patients with neuroblastic tumors was used to validate the performance of a trained Machine Learning (ML) tool to identify and delineate primary neuroblastoma tumors. The dataset was heterogeneous and completely independent from the one used to train and tune the model, consisting of 300 children with neuroblastic tumors having 535 MR T2-weighted sequences (486 sequences at diagnosis and 49 after finalization of the first phase of chemotherapy). The automatic segmentation algorithm was based on a nnU-Net architecture developed within the PRIMAGE project. For comparison, the segmentation masks were manually edited by an expert radiologist, and the time for the manual editing was recorded. Different overlaps and spatial metrics were calculated to compare both masks. The median Dice Similarity Coefficient (DSC) was high 0.997; 0.944-1.000 (median; Q1-Q3). In 18 MR sequences (6%), the net was not able neither to identify nor segment the tumor. No differences were found regarding the MR magnetic field, type of T2 sequence, or tumor location. No significant differences in the performance of the net were found in patients with an MR performed after chemotherapy. The time for visual inspection of the generated masks was 7.9 ± 7.5 (mean ± Standard Deviation (SD)) seconds. Those cases where manual editing was needed (136 masks) required 124 ± 120 s. The automatic CNN was able to locate and segment the primary tumor on the T2-weighted images in 94% of cases. There was an extremely high agreement between the automatic tool and the manually edited masks. This is the first study to validate an automatic segmentation model for neuroblastic tumor identification and segmentation with body MR images. The semi-automatic approach with minor manual editing of the deep learning segmentation increases the radiologist's confidence in the solution with a minor workload for the radiologist.

A comparative study of two automated solutions for cross-sectional skeletal muscle measurement from abdominal computed tomography images.

Measurement of cross-sectional muscle area (CSMA) at the mid third lumbar vertebra (L3) level from computed tomography (CT) images is becoming one of the reference methods for sarcopenia diagnosis. However, manual skeletal muscle segmentation is tedious and is thus restricted to research. Automated solutions are required for use in clinical practice. The aim of this study was to compare the reliability of two automated solutions for the measurement of CSMA. We conducted a retrospective analysis of CT images in our hospital database. We included consecutive individuals hospitalized at the Grenoble University Hospital in France between January and May 2018 with abdominal CT images and sagittal reconstruction. We used two types of software to automatically segment skeletal muscle: ABACS, a module of the SliceOmatic software solution "ABACS-SliceOmatic," and a deep learning-based solution called "AutoMATiCA." Manual segmentation was performed by a medical expert to generate reference data using "SliceOmatic." The Dice similarity coefficient (DSC) was used to measure overlap between the results of the manual and the automated segmentations. The DSC value for each method was compared with the Mann-Whitney U test. A total of 676 hospitalized individuals was retrospectively included (365 males [53.8%] and 312 females [46.2%]). The median DSC for SliceOmatic vs AutoMATiCA (0.969 [5th percentile: 0.909]) was greater than the median DSC for SliceOmatic vs. ABACS-SliceOmatic (0.949 [5th percentile: 0.836]) (p<0.001). AutoMATiCA, which used artificial intelligence, was more reliable than ABACS-SliceOmatic for skeletal muscle segmentation at the L3 level in a cohort of hospitalized individuals. The next step is to develop and validate a neural network that can identify L3 slices, which is currently a fastidious process.

Open-source, fully-automated hybrid cardiac substructure segmentation: development and optimisation

Radiotherapy for thoracic and breast tumours is associated with a range of cardiotoxicities. Emerging evidence suggests cardiac substructure doses may be more predictive of specific outcomes, however, quantitative data necessary to develop clinical planning constraints is lacking. Retrospective analysis of patient data is required, which relies on accurate segmentation of cardiac substructures. In this study, a novel model was designed to deliver reliable, accurate, and anatomically consistent segmentation of 18 cardiac substructures on computed tomography (CT) scans. Thirty manually contoured CT scans were included. The proposed multi-stage method leverages deep learning (DL), multi-atlas mapping, and geometric modelling to automatically segment the whole heart, cardiac chambers, great vessels, heart valves, coronary arteries, and conduction nodes. Segmentation performance was evaluated using the Dice similarity coefficient (DSC), mean distance to agreement (MDA), Hausdorff distance (HD), and volume ratio. Performance was reliable, with no errors observed and acceptable variation in accuracy between cases, including in challenging cases with imaging artefacts and atypical patient anatomy. The median DSC range was 0.81–0.93 for whole heart and cardiac chambers, 0.43–0.76 for great vessels and conduction nodes, and 0.22–0.53 for heart valves. For all structures the median MDA was below 6 mm, median HD ranged 7.7–19.7 mm, and median volume ratio was close to one (0.95–1.49) for all structures except the left main coronary artery (2.07). The fully automatic algorithm takes between 9 and 23 min per case. The proposed fully-automatic method accurately delineates cardiac substructures on radiotherapy planning CT scans. Robust and anatomically consistent segmentations, particularly for smaller structures, represents a major advantage of the proposed segmentation approach. The open-source software will facilitate more precise evaluation of cardiac doses and risks from available clinical datasets.Graphical abstract

Spatial accuracy of computed tomography perfusion to estimate the follow-up infarct on diffusion-weighted imaging after successful mechanical thrombectomy

BackgroundVolumetric accuracy of using computed tomography perfusion (CTP) to estimate the post-treatment infarct in stroke patients with successful recanalization after mechanical thrombectomy (MT) has been studied a lot, however the spatial accuracy and its influence factors has not been fully investigated.MethodsThis retrospective study reviewed the data from consecutive anterior large vessel occlusion (LVO) patients who had baseline CTP, successful recanalization after MT, and post-treatment diffusion-weighed imaging (DWI). Ischemic core on baseline CTP was estimated using relative cerebral blood flood (CBF) of < 30%. The infarct area was outlined manually on post-treatment DWI, and registered to CTP. Spatial agreement was assessed using the Dice similarity coefficient (DSC) and average Hausdorff distance. According to the median DSC, the study population was dichotomized into high and low Dice groups. Univariable and multivariable regression analyses were used to determine the factors independently associated with the spatial agreement.ResultsIn 72 included patients, the median DSC was 0.26, and the median average Hausdorff distance was 1.77 mm. High Dice group showed significantly higher median ischemic core volume on baseline CTP (33.90 mL vs 3.40 mL, P < 0.001), lower proportion of moderate or severe leukoaraiosis [27.78% vs 52.78%, P = 0.031], and higher median infarct volume on follow-up DWI (51.17 mL vs 9.42 mL, P < 0.001) than low Dice group. Ischemic core volume on baseline CTP was found to be independently associated with the spatial agreement (OR, 1.092; P < 0.001).ConclusionsCTP could help to spatially locate the post-treatment infarct in anterior LVO patients who achieving successful recanalization after MT. Ischemic core volume on baseline CTP was independently associated with the spatial agreement.

Automatic segmentation of bladder cancer on MRI using a convolutional neural network and reproducibility of radiomics features: a two-center study

This study aimed to develop a versatile automatic segmentation model of bladder cancer (BC) on MRI using a convolutional neural network and investigate the robustness of radiomics features automatically extracted from apparent diffusion coefficient (ADC) maps. This two-center retrospective study used multi-vendor MR units and included 170 patients with BC, of whom 140 were assigned to training datasets for the modified U-net model with five-fold cross-validation and 30 to test datasets for assessment of segmentation performance and reproducibility of automatically extracted radiomics features. For model input data, diffusion-weighted images with b = 0 and 1000 s/mm2, ADC maps, and multi-sequence images (b0-b1000-ADC maps) were used. Segmentation accuracy was compared between ours and existing models. The reproducibility of radiomics features on ADC maps was evaluated using intraclass correlation coefficient. The model with multi-sequence images achieved the highest Dice similarity coefficient (DSC) with five-fold cross-validation (mean DSC = 0.83 and 0.79 for the training and validation datasets, respectively). The median (interquartile range) DSC of the test dataset model was 0.81 (0.70–0.88). Radiomics features extracted from manually and automatically segmented BC exhibited good reproducibility. Thus, our U-net model performed highly accurate segmentation of BC, and radiomics features extracted from the automatic segmentation results exhibited high reproducibility.

Evaluation of therapeutic radiographer contouring for magnetic resonance image guided online adaptive prostate radiotherapy

Background and PurposeThe implementation of MRI-guided online adaptive radiotherapy has facilitated the extension of therapeutic radiographers’ roles to include contouring, thus releasing the clinician from attending daily treatment. Following undergoing a specifically designed training programme, an online interobserver variability study was performed. Materials and Methods117 images from six patients treated on a MR Linac were contoured online by either radiographer or clinician and the same images contoured offline by the alternate profession. Dice similarity coefficient (DSC), mean distance to agreement (MDA), Hausdorff distance (HD) and volume metrics were used to analyse contours. Additionally, the online radiographer contours and optimised plans (n = 59) were analysed using the offline clinician defined contours. After clinical implementation of radiographer contouring, target volume comparison and dose analysis was performed on 20 contours from five patients. ResultsComparison of the radiographers’ and clinicians’ contours resulted in a median (range) DSC of 0.92 (0.86 – 0.99), median (range) MDA of 0.98 mm (0.2–1.7) and median (range) HD of 6.3 mm (2.5–11.5) for all 117 fractions. There was no significant difference in volume size between the two groups. Of the 59 plans created with radiographer online contours and overlaid with clinicians’ offline contours, 39 met mandatory dose constraints and 12 were acceptable because 95 % of the high dose PTV was covered by 95 % dose, or the high dose PTV was within 3 % of online plan. A clinician blindly reviewed the eight remaining fractions and, using trial quality assurance metrics, deemed all to be acceptable.Following clinical implementation of radiographer contouring, the median (range) DSC of CTV was 0.93 (0.88–1.0), median (range) MDA was 0.8 mm (0.04–1.18) and HD was 5.15 mm (2.09–8.54) respectively. Of the 20 plans created using radiographer online contours overlaid with clinicians’ offline contours, 18 met the dosimetric success criteria, the remaining 2 were deemed acceptable by a clinician. ConclusionRadiographer and clinician prostate and seminal vesicle contours on MRI for an online adaptive workflow are comparable and produce clinically acceptable plans. Radiographer contouring for prostate treatment on a MR-linac can be effectively introduced with appropriate training and evaluation. A DSC threshold for target structures could be implemented to streamline future training.

Interobserver Variations in Target Delineation in Intensity-Modulated Radiation Therapy for Nasopharyngeal Carcinoma and its Impact on Target Dose Coverage.

To investigate the differences between physicians in target delineation in intensity-modulated radiation therapy for nasopharyngeal carcinoma as well as their impact on target dose coverage. Ninety-nine in-hospital patients were randomly selected for retrospective analysis, and the target volumes were delineated by 2 physicians. The target volumes were integrated with the original plans, and the differential parameters, including the Dice similarity coefficient (DSC), Hausdorff distance (HD), and Jaccard similarity coefficient (JSC) were recorded. The dose-volume parameters to evaluate target dose coverage were analyzed by superimposing the same original plan to the 2 sets of images on which the target volumes were contoured by the 2 physicians. The significance of differences in target volumes and dose coverage were evaluated using statistical analysis. The target dose coverage for different sets of target volumes showed statistically significant differences, while the similarity metrics to evaluate geometric target volume differences did not. More specifically, for PGTVnx, the median DSC, JSC, and HD were 0.85, 0.74, and 11.73, respectively; for PCTV1, the median values were 0.87, 0.77, and 11.78, respectively; for PCTV2, the median values were 0.90, 0.82, and 16.12, respectively. For patients in stages T3-4, DSC, and JSC were reduced but HD was increased compared to those in stages T1-2. Dosimetric analysis indicated that, for the target volumes, significant differences between the 2 physicians were found in D95, D99, and V100 for all the target volumes (ie, PGTVnx, PCTV1, and PCTV2) across the whole group of patients, as well as in patients with disease stages T3-4 and T1-2. The target volumes delineated by the 2 physicians had a high similarity, but the maximal distances between the outer contours of the 2 sets were significantly different. In patients with advanced T stages, significant differences in dose distributions were found, stemming from the deviations of target delineation.

Inter-observer variation of target volume delineation for CT-guided cervical cancer brachytherapy.

Delineation is a critical and challenging step in radiotherapy planning. Differences in delineation among observers are common, despite the presence of contouring guidelines. This study aimed to identify the inter-observer variability in the target volume delineation of computed tomography (CT)-guided brachytherapy for cervical cancer. Four radiation oncologists (ROs) with different expertise levels delineated high-risk (HR) and intermediate-risk (IR) clinical target volume (CTV) according to GYN GEC-ESTRO recommendations, in a blinded manner on every CT set of ten locally advanced cervical cancer cases. The most experienced RO's contours were determined as the index and used for comparison. Dice similarity coefficient (DSC) and pairwise Hausdorff distance (HD) metrics were applied to compare the overlap and gross deviations of all contours. Median DSC for HR-CTV and IR-CTV were 0.73 and 0.76, respectively, and a good concordance was achieved for both in majority of contours. While there was no difference in DSC measurements for HR-CTV among the three ROs, RO-3 provided improved DSC values for IR-CTV (p = 0.01). Median HD95 was 5.02 mm and 6.83 mm, and median HDave was 1.69 mm and 2.21 mm for HR-CTV and IR-CTV, respectively. There was no significant difference among ROs in HR-CTV for HD95 or HDave; however, IR-CTV value was significantly improved according to RO-3 (p = 0.01). Case-by-case HD analysis showed no significant inter-observer variations, except for two cases. The inter-observer agreement is generally high for target volumes in CT-guided brachytherapy for cervical cancer. The agreement is lower for IR-CTV than HR-CTV. The individual characteristics of each case and different expertise levels of the ROs may have caused the differences. Despite the good concordance for delineation, dosimetric consequences can still be clinically significant.

Quantification of Epicardial Adipose Tissue Volume and Attenuation for Cardiac CT Scans Using Deep Learning in a Single Multi-Task Framework.

Recent studies have shown that epicardial adipose tissue (EAT) is an independent atrial fibrillation (AF) prognostic marker and has influence on the myocardial function. In computed tomography (CT), EAT volume (EATv) and density (EATd) are parameters that are often used to quantify EAT. While increased EATv has been found to correlate with the prevalence and the recurrence of AF after ablation therapy, higher EATd correlates with inflammation due to arrest of lipid maturation and with high risk of plaque presence and plaque progression. Automation of the quantification task diminishes the variability in readings introduced by different observers in manual quantification and results in high reproducibility of studies and less time-consuming analysis. Our objective is to develop a fully automated quantification of EATv and EATd using a deep learning (DL) framework. We proposed a framework that consists of image classification and segmentation DL models and performs the task of selecting images with EAT from all the CT images acquired for a patient, and the task of segmenting the EAT from the output images of the preceding task. EATv and EATd are estimated using the segmentation masks to define the region of interest. For our experiments, a 300-patient dataset was divided into two subsets, each consisting of 150 patients: Dataset 1 (41,979 CT slices) for training the DL models, and Dataset 2 (36,428 CT slices) for evaluating the quantification of EATv and EATd. The classification model achieved accuracies of 98% for precision, recall and scores, and the segmentation model achieved accuracies in terms of mean ( std.) and median dice similarity coefficient scores of 0.844 ( 0.19) and 0.84, respectively. Using the evaluation set (Dataset 2), our approach resulted in a Pearson correlation coefficient of 0.971 ( = 0.943) between the label and predicted EATv, and the correlation coefficient of 0.972 ( = 0.945) between the label and predicted EATd. We proposed a framework that provides a fast and robust strategy for accurate EAT segmentation, and volume (EATv) and attenuation (EATd) quantification tasks. The framework will be useful to clinicians and other practitioners for carrying out reproducible EAT quantification at patient level or for large cohorts and high-throughput projects.

NS-HGlio: A generalizable and repeatable HGG segmentation and volumetric measurement AI algorithm for the longitudinal MRI assessment to inform RANO in trials and clinics.

Accurate and repeatable measurement of high-grade glioma (HGG) enhancing (Enh.) and T2/FLAIR hyperintensity/edema (Ed.) is required for monitoring treatment response. 3D measurements can be used to inform the modified Response Assessment in Neuro-oncology criteria. We aim to develop an HGG volumetric measurement and visualization AI algorithm that is generalizable and repeatable. A single 3D-Convoluted Neural Network, NS-HGlio, to analyze HGG on MRIs using 5-fold cross validation was developed using retrospective (557 MRIs), multicentre (38 sites) and multivendor (32 scanners) dataset divided into training (70%), validation (20%), and testing (10%). Six neuroradiologists created the ground truth (GT). Additional Internal validation (IV, three institutions) using 70 MRIs, and External validation (EV, single institution) using 40 MRIs through measuring the Dice Similarity Coefficient (DSC) of Enh., Ed. ,and Enh. + Ed. (WholeLesion/WL) tumor tissue and repeatability testing on 14 subjects from the TCIA MGH-QIN-GBM dataset using volume correlations between timepoints were performed. IV Preoperative median DSC Enh. 0.89 (SD 0.11), Ed. 0.88 (0.28), WL 0.88 (0.11). EV Preoperative median DSC Enh. 0.82 (0.09), Ed. 0.83 (0.11), WL 0.86 (0.06). IV Postoperative median DSC Enh. 0.77 (SD 0.20), Ed 0.78. (SD 0.09), WL 0.78 (SD 0.11). EV Postoperative median DSC Enh. 0.75 (0.21), Ed 0.74 (0.12), WL 0.79 (0.07). Repeatability testing; Intraclass Correlation Coefficient of 0.95 Enh. and 0.92 Ed. NS-HGlio is accurate, repeatable, and generalizable. The output can be used for visualization, documentation, treatment response monitoring, radiation planning, intra-operative targeting, and estimation of Residual Tumor Volume among others.

Repeatability of MRI for radiotherapy planning for pelvic, brain, and head and neck malignancies

Objectives: The objective of this study was to assess the repeatability of MRI for the purpose of radiotherapy treatment planning by considering the difference in registering MRI to MRI compared to registering CT and MRI.Methods: Fifty patients undergoing radiotherapy planning for gynaecological, prostate, rectal, head and neck and CNS malignancies had a planning CT followed by two T2-weighted MRIs. Anatomical landmarks were contoured on each dataset and the images were rigidly registered. Centre of Mass (COM), Dice Similarity Coefficient (DSC), and Mean Distance to Agreement (MDA) were calculated to assess structure volume and position comparing CT-MRI and MRI-MRI.Results: DSC and MDA demonstrated more consistency in delineated volumes for MRI-MRI than for the CT-MRI comparison. The median DSC values were ≥0.8 for 15 of 46 contoured structures for the CT-MRI comparison and 21 of 23 structures for the MRI-MRI comparison. MDA values were ≤1 mm for 11 of 46 structures for the CT-MRI comparison and 18 of 23 structures for the MRI-MRI comparisons. COM were also more consistent between MRI-MRI than between CT-MRI.Conclusion: There was less variability of anatomical structures between repeated MRIs than registered CT and MRI datasets, demonstrating consistency of MRI for volume delineation in radiotherapy planning.

Predicting final ischemic stroke lesions from initial diffusion-weighted images using a deep neural network

BackgroundFor prognosis of stroke, measurement of the diffusion-perfusion mismatch is a common practice for estimating tissue at risk of infarction in the absence of timely reperfusion. However, perfusion-weighted imaging (PWI) adds time and expense to the acute stroke imaging workup. We explored whether a deep convolutional neural network (DCNN) model trained with diffusion-weighted imaging obtained at admission could predict final infarct volume and location in acute stroke patients. MethodsIn 445 patients, we trained and validated an attention-gated (AG) DCNN to predict final infarcts as delineated on follow-up studies obtained 3 to 7 days after stroke. The input channels consisted of MR diffusion-weighted imaging (DWI), apparent diffusion coefficients (ADC) maps, and thresholded ADC maps with values less than 620 × 10−6 mm2/s, while the output was a voxel-by-voxel probability map of tissue infarction. We evaluated performance of the model using the area under the receiver-operator characteristic curve (AUC), the Dice similarity coefficient (DSC), absolute lesion volume error, and the concordance correlation coefficient (ρc) of the predicted and true infarct volumes. ResultsThe model obtained a median AUC of 0.91 (IQR: 0.84–0.96). After thresholding at an infarction probability of 0.5, the median sensitivity and specificity were 0.60 (IQR: 0.16–0.84) and 0.97 (IQR: 0.93–0.99), respectively, while the median DSC and absolute volume error were 0.50 (IQR: 0.17–0.66) and 27 ml (IQR: 7–60 ml), respectively. The model’s predicted lesion volumes showed high correlation with ground truth volumes (ρc = 0.73, p < 0.01). ConclusionAn AG-DCNN using diffusion information alone upon admission was able to predict infarct volumes at 3–7 days after stroke onset with comparable accuracy to models that consider both DWI and PWI. This may enable treatment decisions to be made with shorter stroke imaging protocols.

A Dual-Channel Deep Learning Approach for Lung Cavity Estimation From Hyperpolarized Gas and Proton MRI.

Hyperpolarized gas MRI can quantify regional lung ventilation via biomarkers, including the ventilation defect percentage (VDP). VDP is computed from segmentations derived from spatially co-registered functional hyperpolarized gas and structural proton (1 H)-MRI. Although acquired at similar lung inflation levels, they are frequently misaligned, requiring a lung cavity estimation (LCE). Recently, single-channel, mono-modal deep learning (DL)-based methods have shown promise for pulmonary image segmentation problems. Multichannel, multimodal approaches may outperform single-channel alternatives. We hypothesized that a DL-based dual-channel approach, leveraging both 1 H-MRI and Xenon-129-MRI (129 Xe-MRI), can generate LCEs more accurately than single-channel alternatives. Retrospective. A total of 480 corresponding 1 H-MRI and 129 Xe-MRI scans from 26 healthy participants (median age [range]: 11 [8-71]; 50% females) and 289 patients with pulmonary pathologies (median age [range]: 47 [6-83]; 51% females) were split into training (422 scans [88%]; 257 participants [82%]) and testing (58 scans [12%]; 58 participants [18%]) sets. 1.5-T, three-dimensional (3D) spoiled gradient-recalled 1 H-MRI and 3D steady-state free-precession 129 Xe-MRI. We developed a multimodal DL approach, integrating 129 Xe-MRI and 1 H-MRI, in a dual-channel convolutional neural network. We compared this approach to single-channel alternatives using manually edited LCEs as a benchmark. We further assessed a fully automatic DL-based framework to calculate VDPs and compared it to manually generated VDPs. Friedman tests with post hoc Bonferroni correction for multiple comparisons compared single-channel and dual-channel DL approaches using Dice similarity coefficient (DSC), average boundary Hausdorff distance (average HD), and relative error (XOR) metrics. Bland-Altman analysis and paired t-tests compared manual and DL-generated VDPs. A P value < 0.05 was considered statistically significant. The dual-channel approach significantly outperformed single-channel approaches, achieving a median (range) DSC, average HD, and XOR of 0.967 (0.867-0.978), 1.68 mm (37.0-0.778), and 0.066 (0.246-0.045), respectively. DL-generated VDPs were statistically indistinguishable from manually generated VDPs (P= 0.710). Our dual-channel approach generated LCEs, which could be integrated with ventilated lung segmentations to produce biomarkers such as the VDP without manual intervention. 4. Stage 1.

NIMG-75. REPEATABILITY OF MANUAL SEGMENTATION OF GLIOBLASTOMA ON MRI - QUALITY ASSURANCE FOR A QUANTITATIVE MRI RADIOMICS REPEATABILITY STUDY

Abstract BACKGROUND Accurate and repeatable imaging segmentation of glioblastomas is important for effective radiomics modelling of clinical endpoints such as survival and treatment response. Manual and even automated segmentation are prone to variability. The primary objective of this study is to evaluate the repeatability of manually segmented glioblastomas on MRI which will serve as quality assurance for a quantitative MRI radiomics repeatability study. METHODS MRI datasets from n=19 patients with recurrent glioblastoma extracted from The Cancer Imaging Archive (RIDER Neuro MRI dataset) comprised of test-retest scans (time-point 1 and 2) acquired on average 2 days apart. The MRI protocol consisted of dynamic contrast-enhanced 3D FLASH using 0.1mmol/kg Magnevist intravenous injection at 3cc/second (TR 3.8ms, TE 1.8ms, 5mm slices, flip angle 25°), contrast-enhanced 3D FLASH (TR 8.6ms, TE 4.1ms, 1mm slices, flip angle 20°) and contrast-enhanced 3D FLAIR (TR 6000ms, TE 353ms, TI 2200ms, 1mm slices, flip angle 180°) on a 1.5T magnet. Contrast enhancing tumour from n=38 MRI scans were manually segmented by a radiologist slice by slice using contrast-enhanced 3D FLASH T1-weighted images on MIM software (v6.9.5). All cases at time-point 1 were contoured before contouring commenced for time-point 2 in the same order. Repeatability and spatial overlap was assessed by Dice similarity coefficient (DSC), Hausdorff distance (HD) and centroid shift. RESULTS A comparison of time-point 1 and 2 demonstrated the median DSC for n=19 cases was 0.84 (interquartile range 0.07), mean HD was only 0.1cm and mean centroid shift was 0.2cm. CONCLUSION High repeatability and spatial overlap of manually segmented contrast enhancing regions of recurrent glioblastoma were achieved on serial MRI. This substantiates a high level of repeatability which forms a core component of an MRI radiomics platform under development.

Patient-specific daily updated deep learning auto-segmentation for MRI-guided adaptive radiotherapy

Background and purposeDeep Learning (DL) technique has shown great potential but still has limited success in online contouring for MR-guided adaptive radiotherapy (MRgART). This study proposed a patient-specific DL auto-segmentation (DLAS) strategy using the patient’s previous images and contours to update the model and improve segmentation accuracy and efficiency for MRgART. Methods and materialsA prototype model was trained for each patient using the first set of MRI and corresponding contours as inputs. The patient-specific model was updated after each fraction with all the available fractional MRIs/contours, and then used to predict the segmentation for the next fraction. During model training, a variant was fitted under consistency constraints, limiting the differences in the volume, length and centroid between the predictions for the latest MRI within a reasonable range. The model performance was evaluated for both organ-at-risks and tumors auto-segmentation for a total of 6 abdominal/pelvic cases (each with at least 8 sets of MRIs/contours) underwent MRgART through Dice Similarity Coefficient (DSC) and 95% Hausdorff Distance (HD95), and was compared with deformable image registration (DIR) and frozen DL model (no updating after pre-training). The contouring time was also recorded and analyzed. ResultsThe proposed model achieved superior performance with higher mean DSC (0.90, 95 % CI: 0.88–0.95), as compared to DIR (0.63, 95 %CI: 0.59–0.68) and frozen DL models (0.74, 95 % CI: 0.71–0.79). As for tumors, the proposed method yielded a median DSC of 0.95, 95 % CI: 0.94–0.97, and a median HD95 of 1.63 mm, 95 % CI: 1.22 mm-2.06 mm. The contouring time was reduced significantly (p < 0.05) using the proposed method (73.4 ± 6.5 secs) compared to the manual process (12 ∼ 22 mins). The online ART time was reduced to 1650 ± 274 seconds with the proposed method, as compared to 3251.8 ± 447 seconds using the original workflow. ConclusionThe proposed patient-specific DLAS method can significantly improve the segmentation accuracy and efficiency for longitudinal MRIs, thereby facilitating the routine practice of MRgART.

Semi-automatic segmentation of the fetal brain from magnetic resonance imaging.

Volumetric measurements of fetal brain maturation in the third trimester of pregnancy are key predictors of developmental outcomes. Improved understanding of fetal brain development trajectories may aid in identifying and clinically managing at-risk fetuses. Currently, fetal brain structures in magnetic resonance images (MRI) are often manually segmented, which requires both time and expertise. To facilitate the targeting and measurement of brain structures in the fetus, we compared the results of five segmentation methods applied to fetal brain MRI data to gold-standard manual tracings. Adult women with singleton pregnancies (n = 21), of whom five were scanned twice, approximately 3 weeks apart, were recruited [26 total datasets, median gestational age (GA) = 34.8, IQR = 30.9-36.6]. T2-weighted single-shot fast spin echo images of the fetal brain were acquired on 1.5T and 3T MRI scanners. Images were first combined into a single 3D anatomical volume. Next, a trained tracer manually segmented the thalamus, cerebellum, and total cerebral volumes. The manual segmentations were compared with five automatic methods of segmentation available within Advanced Normalization Tools (ANTs) and FMRIB's Linear Image Registration Tool (FLIRT) toolboxes. The manual and automatic labels were compared using Dice similarity coefficients (DSCs). The DSC values were compared using Friedman's test for repeated measures. Comparing cerebellum and thalamus masks against the manually segmented masks, the median DSC values for ANTs and FLIRT were 0.72 [interquartile range (IQR) = 0.6-0.8] and 0.54 (IQR = 0.4-0.6), respectively. A Friedman's test indicated that the ANTs registration methods, primarily nonlinear methods, performed better than FLIRT (p < 0.001). Deformable registration methods provided the most accurate results relative to manual segmentation. Overall, this semi-automatic subcortical segmentation method provides reliable performance to segment subcortical volumes in fetal MR images. This method reduces the costs of manual segmentation, facilitating the measurement of typical and atypical fetal brain development.

Deep Learning for Automated Elective Lymph Node Level Segmentation for Head and Neck Cancer Radiotherapy.

Depending on the clinical situation, different combinations of lymph node (LN) levels define the elective LN target volume in head-and-neck cancer (HNC) radiotherapy. The accurate auto-contouring of individual LN levels could reduce the burden and variability of manual segmentation and be used regardless of the primary tumor location. We evaluated three deep learning approaches for the segmenting individual LN levels I-V, which were manually contoured on CT scans from 70 HNC patients. The networks were trained and evaluated using five-fold cross-validation and ensemble learning for 60 patients with (1) 3D patch-based UNets, (2) multi-view (MV) voxel classification networks and (3) sequential UNet+MV. The performances were evaluated using Dice similarity coefficients (DSC) for automated and manual segmentations for individual levels, and the planning target volumes were extrapolated from the combined levels I-V and II-IV, both for the cross-validation and for an independent test set of 10 patients. The median DSC were 0.80, 0.66 and 0.82 for UNet, MV and UNet+MV, respectively. Overall, UNet+MV significantly (p &lt; 0.0001) outperformed other arrangements and yielded DSC = 0.87, 0.85, 0.86, 0.82, 0.77, 0.77 for the combined and individual level I-V structures, respectively. Both PTVs were also significantly (p &lt; 0.0001) more accurate with UNet+MV, with DSC = 0.91 and 0.90, respectively. The accurate segmentation of individual LN levels I-V can be achieved using an ensemble of UNets. UNet+MV can further refine this result.

Dosimetric Implications of Margin-Reduced MRI-Guided Stereotactic Body Radiotherapy to the Prostate Bed Following Radical Prostatectomy: Post-Hoc Analysis of a Phase II Study

<h3>Purpose/Objective(s)</h3> Due to the highly deformable and mobile nature of the target volume and adjacent organs at risk (OARs), generous planning target volume (PTV) margins have historically been used in the context of radiotherapy after radical prostatectomy. MRI-guided radiotherapy systems have improved on-board imaging capabilities and the ability to perform automatic beam holds based on pre-specified structures. Herein, we evaluate the dosimetric implications of interfraction variations in the clinical target volumes (CTV) and OARs in patients receiving margin-reduced MRI-guided stereotactic body radiotherapy (SBRT) to the prostate bed. <h3>Materials/Methods</h3> 31 consecutive patients received MRI-guided SBRT to the prostate bed (30 – 34 Gy in 5 fractions) without adaptive planning on a phase II study. The OAR volumes, CTV, and PTV (created using a 3 mm isotropic expansion of the CTV) were retrospectively contoured on on-board 0.35T MRIs obtained prior to each fraction (n=155). Geometric comparisons were made between the initial planning contours and the 5 daily contours for each patient using change in volume and Dice similarity coefficient (DSC) calculations. Coverage of the CTV and PTV was assessed (CTV V95% >93% and PTV V95% >90%), as were bladder and rectal dose constraints (Bladder Dmax <36.7 Gy, Bladder V32.5 Gy <35%, Rectum Dmax <36.7 Gy, Rectum V27.5 Gy <45%, Rectum 32.5 Gy <30%, and Rectal Wall V24 <50%). <h3>Results</h3> The median change in volume was -24.5% (interquartile range [IQR], -35.5 – -2.2%) for the bladder, 6.95% (IQR, -6.94 – 23.0%) for the rectum, -0.2% (IQR, -8.3 – 8.7%) for the rectal wall, 0.48% (IQR, -3.87 – 7.6%) for the CTV, and -0.48% (IQR, -5.12 – 6.53%) for the PTV. The median DSC was 0.79 (IQR, 0.73 – 0.84) for the bladder, 0.76 (IQR, 0.69 – 0.81) for the rectum, 0.35 (IQR, 0.28 – 0.44) for the rectal wall, 0.89 (IQR, 0.83 – 0.93) for the CTV, and 0.90 (IQR, 0.87 – 0.95) for the PTV. 145/155 fractions (93.5%) met the CTV V95% >93% target and 134/155 fractions (86.5%) met both the CTV V95% >93% and PTV V95% >90% targets. When calculated over 5 cumulative fractions, 29/31 patients (93.5%) met the CTV goal of V95% >93%. 62/155 fractions (40%) failed at least one OAR constraint. The mean number of fractions per patient that failed at least one OAR constraint was 2/5 (median, 1; range, 0 – 5). <h3>Conclusion</h3> With MRI-guidance, a reduction of the PTV margin to 3 mm isotropically allows for adequate CTV coverage in over 90% of patients without the need for adaptive planning. However, OAR constraints are not met in 40% of delivered fractions despite meeting all dose constraints at the time of treatment planning. Online dose prediction and, if indicated, adaptive radiotherapy can optimize this to allow for improved OAR protection and potentially allow for further reductions in the PTV expansion while maintaining adequate CTV coverage.